scikit-learn · glemaitre · Feb 13, 2017 · Feb 14, 2017 · Feb 14, 2017 · Feb 14, 2017
diff --git a/doc/modules/linear_model.rst b/doc/modules/linear_model.rst
@@ -1222,7 +1222,7 @@ polynomial regression can be created and used as follows::
     >>> x = np.arange(5)
     >>> y = 3 - 2 * x + x ** 2 - x ** 3
     >>> model = model.fit(x[:, np.newaxis], y)
-    >>> model.named_steps['linear'].coef_
+    >>> model.named_steps_['linear'].coef_
     array([ 3., -2.,  1., -1.])
 
 The linear model trained on polynomial features is able to exactly recover

diff --git a/doc/modules/pipeline.rst b/doc/modules/pipeline.rst
@@ -50,16 +50,13 @@ it takes a variable number of estimators and returns a pipeline,
 filling in the names automatically::
 
     >>> from sklearn.pipeline import make_pipeline
-    >>> from sklearn.naive_bayes import MultinomialNB
-    >>> from sklearn.preprocessing import Binarizer
-    >>> make_pipeline(Binarizer(), MultinomialNB()) # doctest: +NORMALIZE_WHITESPACE
+    >>> make_pipeline(PCA(), SVC()) # doctest: +NORMALIZE_WHITESPACE, +ELLIPSIS
     Pipeline(memory=None,
-             steps=[('binarizer', Binarizer(copy=True, threshold=0.0)),
-                    ('multinomialnb', MultinomialNB(alpha=1.0,
-                                                    class_prior=None,
-                                                    fit_prior=True))])
+             steps=[('pca', PCA(copy=True,...)),
+                    ('svc', SVC(C=1.0,...))])
 
-The estimators of a pipeline are stored as a list in the ``steps`` attribute::
+The original estimators of a pipeline are stored as a list in the ``steps``
+attribute::
 
     >>> pipe.steps[0]
     ('reduce_dim', PCA(copy=True, iterated_power='auto', n_components=None, random_state=None,
@@ -71,6 +68,23 @@ and as a ``dict`` in ``named_steps``::
     PCA(copy=True, iterated_power='auto', n_components=None, random_state=None,
       svd_solver='auto', tol=0.0, whiten=False)
 
+Once the pipeline has been fitted, ``steps_`` and ``named_steps_`` have to be
+used.
+
+    >>> from sklearn.datasets import load_iris
+    >>> iris = load_iris()
+    >>> pipe.fit(iris.data, iris.target)
+    ... # doctest: +NORMALIZE_WHITESPACE, +ELLIPSIS
+    Pipeline(memory=None,
+             steps=[('reduce_dim', PCA(copy=True,...)),
+                    ('clf', SVC(C=1.0,...))])
+    >>> pipe.named_steps_['reduce_dim']  # doctest: +NORMALIZE_WHITESPACE
+    PCA(copy=True, iterated_power='auto', n_components=None, random_state=None,
+        svd_solver='auto', tol=0.0, whiten=False)
+    >>> pipe.steps_[0]  # doctest: +NORMALIZE_WHITESPACE
+    ('reduce_dim', PCA(copy=True, iterated_power='auto', n_components=None,
+     random_state=None, svd_solver='auto', tol=0.0, whiten=False))
+
 Parameters of the estimators in the pipeline can be accessed using the
 ``<estimator>__<parameter>`` syntax::
 
@@ -152,48 +166,6 @@ object::
     >>> # Clear the cache directory when you don't need it anymore
     >>> rmtree(cachedir)
 
-.. warning:: **Side effect of caching transfomers**
-
-   Using a :class:`Pipeline` without cache enabled, it is possible to
-   inspect the original instance such as::
-
-     >>> from sklearn.datasets import load_digits
-     >>> digits = load_digits()
-     >>> pca1 = PCA()
-     >>> svm1 = SVC()
-     >>> pipe = Pipeline([('reduce_dim', pca1), ('clf', svm1)])
-     >>> pipe.fit(digits.data, digits.target)
-     ... # doctest: +NORMALIZE_WHITESPACE, +ELLIPSIS
-     Pipeline(memory=None,
-              steps=[('reduce_dim', PCA(...)), ('clf', SVC(...))])
-     >>> # The pca instance can be inspected directly
-     >>> print(pca1.components_) # doctest: +NORMALIZE_WHITESPACE, +ELLIPSIS
-         [[ -1.77484909e-19  ... 4.07058917e-18]]
-
-   Enabling caching triggers a clone of the transformers before fitting.
-   Therefore, the transformer instance given to the pipeline cannot be
-   inspected directly.
-   In following example, accessing the :class:`PCA` instance ``pca2``
-   will raise an ``AttributeError`` since ``pca2`` will be an unfitted
-   transformer.
-   Instead, use the attribute ``named_steps`` to inspect estimators within
-   the pipeline::
-
-     >>> cachedir = mkdtemp()
-     >>> pca2 = PCA()
-     >>> svm2 = SVC()
-     >>> cached_pipe = Pipeline([('reduce_dim', pca2), ('clf', svm2)],
-     ...                        memory=cachedir)
-     >>> cached_pipe.fit(digits.data, digits.target)
-     ... # doctest: +NORMALIZE_WHITESPACE, +ELLIPSIS
-      Pipeline(memory=...,
-               steps=[('reduce_dim', PCA(...)), ('clf', SVC(...))])
-     >>> print(cached_pipe.named_steps['reduce_dim'].components_)
-     ... # doctest: +NORMALIZE_WHITESPACE, +ELLIPSIS
-         [[ -1.77484909e-19  ... 4.07058917e-18]]
-     >>> # Remove the cache directory
-     >>> rmtree(cachedir)
-
 .. topic:: Examples:
 
  * :ref:`sphx_glr_auto_examples_plot_compare_reduction.py`

diff --git a/doc/whats_new.rst b/doc/whats_new.rst
@@ -56,6 +56,7 @@ Enhancements
    - :class:`multioutput.MultiOutputRegressor` and :class:`multioutput.MultiOutputClassifier`
      now support online learning using `partial_fit`.
      issue: `8053` by :user:`Peng Yu <yupbank>`.
+
    - :class:`pipeline.Pipeline` allows to cache transformers
      within a pipeline by using the ``memory`` constructor parameter.
      By :issue:`7990` by :user:`Guillaume Lemaitre <glemaitre>`.
@@ -251,6 +252,13 @@ API changes summary
       :func:`sklearn.model_selection.cross_val_predict`.
       :issue:`2879` by :user:`Stephen Hoover <stephen-hoover>`.
 
+    - In the future, the estimators in the ``steps`` and ``named_steps``
+      attributes will no longer have their ``fit()`` methods called directly.
+      Users will have to access fitted Pipeline steps in ``steps_``
+      and ``named_steps_`. The warning was introduced in 0.19 and will take
+      effect in 0.22.
+      :issue:`8350` by :user:`Guillaume Lemaitre <glemaitre>`.
+
 .. _changes_0_18_1:
 
 Version 0.18.1

diff --git a/examples/neural_networks/plot_rbm_logistic_classification.py b/examples/neural_networks/plot_rbm_logistic_classification.py
@@ -129,7 +129,7 @@ def nudge_dataset(X, Y):
 # Plotting
 
 plt.figure(figsize=(4.2, 4))
-for i, comp in enumerate(rbm.components_):
+for i, comp in enumerate(classifier.named_steps_['rbm'].components_):
     plt.subplot(10, 10, i + 1)
     plt.imshow(comp.reshape((8, 8)), cmap=plt.cm.gray_r,
                interpolation='nearest')

diff --git a/examples/plot_digits_pipe.py b/examples/plot_digits_pipe.py
@@ -57,7 +57,7 @@
                               logistic__C=Cs))
 estimator.fit(X_digits, y_digits)
 
-plt.axvline(estimator.best_estimator_.named_steps['pca'].n_components,
+plt.axvline(estimator.best_estimator_.named_steps_['pca'].n_components,
             linestyle=':', label='n_components chosen')
 plt.legend(prop=dict(size=12))
 plt.show()
diff --git a/examples/preprocessing/plot_scaling_importance.py b/examples/preprocessing/plot_scaling_importance.py
@@ -86,15 +86,15 @@
 print('{:.2%}\n'.format(metrics.accuracy_score(y_test, pred_test_std)))
 
 # Extract PCA from pipeline
-pca = unscaled_clf.named_steps['pca']
-pca_std = std_clf.named_steps['pca']
+pca = unscaled_clf.named_steps_['pca']
+pca_std = std_clf.named_steps_['pca']
 
 # Show first principal componenets
 print('\nPC 1 without scaling:\n', pca.components_[0])
 print('\nPC 1 with scaling:\n', pca_std.components_[0])
 
 # Scale and use PCA on X_train data for visualization.
-scaler = std_clf.named_steps['standardscaler']
+scaler = std_clf.named_steps_['standardscaler']
 X_train_std = pca_std.transform(scaler.transform(X_train))
 
 # visualize standardized vs. untouched dataset with PCA performed

diff --git a/sklearn/feature_extraction/tests/test_text.py b/sklearn/feature_extraction/tests/test_text.py
@@ -246,7 +246,7 @@ def test_countvectorizer_custom_vocabulary_pipeline():
         ('count', CountVectorizer(vocabulary=what_we_like)),
         ('tfidf', TfidfTransformer())])
     X = pipe.fit_transform(ALL_FOOD_DOCS)
-    assert_equal(set(pipe.named_steps['count'].vocabulary_),
+    assert_equal(set(pipe.named_steps_['count'].vocabulary_),
                  set(what_we_like))
     assert_equal(X.shape[1], len(what_we_like))
 
@@ -728,7 +728,7 @@ def test_count_vectorizer_pipeline_grid_selection():
     # the grid_search is considered the best estimator since they all converge
     # to 100% accuracy models
     assert_equal(grid_search.best_score_, 1.0)
-    best_vectorizer = grid_search.best_estimator_.named_steps['vect']
+    best_vectorizer = grid_search.best_estimator_.named_steps_['vect']
     assert_equal(best_vectorizer.ngram_range, (1, 1))
 
 
@@ -765,7 +765,7 @@ def test_vectorizer_pipeline_grid_selection():
     # the grid_search is considered the best estimator since they all converge
     # to 100% accuracy models
     assert_equal(grid_search.best_score_, 1.0)
-    best_vectorizer = grid_search.best_estimator_.named_steps['vect']
+    best_vectorizer = grid_search.best_estimator_.named_steps_['vect']
     assert_equal(best_vectorizer.ngram_range, (1, 1))
     assert_equal(best_vectorizer.norm, 'l2')
     assert_false(best_vectorizer.fixed_vocabulary_)